An impact load simulation system based on data restoration technology

Abstract

This invention relates to an impact load simulation system based on data restoration technology, comprising: Step 1, preprocessing the collected load data by cleaning, completing, and deleting; Step 2, detecting outliers after preprocessing using the 3σ principle, and correcting the outliers; Step 3, establishing an abnormal load data correction model; and Step 4, analyzing the data results after the correction model. Normalization is used to process abnormal data of different dimensions, and the experimental results are effectively statistically analyzed. A BP neural network is used to establish an abnormal load data correction model, achieving better results through self-learning, thus better correcting the abnormal data and obtaining load prediction values ​​to replace the data outliers.

Description

Technical Field

[0001] This invention relates to the field of data restoration technology, and in particular to an impact load simulation system based on data restoration technology. Background Technology

[0002] In the operation of power systems, various faults can cause data anomalies at different stages of the data acquisition process, resulting in discrepancies between the transmitted data, actual values, and predicted values. Even when the data acquisition system is functioning normally, abnormal load changes can occur due to line maintenance outages, large user loads, or major events. In smart grids, occasional line or equipment faults in metering devices and untimely data backups in the control center's storage devices can also lead to data gaps and anomalies in the acquired load data, causing the observed data to deviate from the norm. All these abnormal observation data are collectively referred to as outliers. These outliers can mislead the establishment of load forecasting models and affect the accuracy of forecast results. Therefore, it is necessary to identify and correct outliers.

[0003] Chinese patent CN 106709816 B discloses a method for identifying and correcting abnormal power load data based on nonparametric regression analysis. This method can identify and correct abnormal power load data, including large industrial power load data, and can also improve the theoretical deficiencies of load anomaly data identification and correction theory in power load data processing. However, the correction method of this patent requires a lot of effort to construct features, and the features constructed for power load are often more extensive, so its applicability is not strong. Summary of the Invention

[0004] This invention provides an impact load simulation system based on data restoration technology, which utilizes data normalization processing and BP neural network to perform self-learning on the data, and to better correct and fit abnormal data.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] An impact load simulation system based on data reconstruction technology includes the following steps:

[0007] Step 1: Preprocess the collected load data;

[0008] Step 2: Process the preprocessed outliers;

[0009] Step 3: Establish an abnormal load data correction model;

[0010] Step 4: Analyze the data results after the model has been corrected.

[0011] Furthermore, step one cleans the collected raw data, fills in the missing values ​​in the data, deletes data with a large amount of missing values ​​directly, fills in the missing data with a small amount of missing values ​​using linear interpolation, and normalizes data of different dimensions.

[0012] Furthermore, in step two, the preprocessed data is subjected to outlier detection using the 3σ principle, and then outlier correction is performed.

[0013] Furthermore, in step three, the preprocessed historical load data is divided into a training set, a test set, and a validation set for the anomaly data correction model, and the BP neural network prediction model is iteratively trained.

[0014] Furthermore, step four uses the mean absolute percentage error (MASE) evaluation index to measure the prediction results; the closer the MSE value is to 0, the better the prediction results.

[0015] Compared with the prior art, the beneficial effects of the present invention are:

[0016] 1) Normalization was used to process outlier data of different dimensions, and the experimental results were effectively statistically analyzed.

[0017] 2) Utilize a BP neural network to establish an abnormal load data correction model, achieving better results through self-learning;

[0018] 3) It achieves better correction of abnormal data, and obtains load forecast values ​​to replace data anomalies. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the present invention.

[0020] Figure 2 This is a flowchart of the BP neural network prediction model described in this invention. Detailed Implementation

[0021] The specific embodiments of the present invention will be further described below with reference to the accompanying drawings:

[0022] See Figure 1 This is a flowchart illustrating the present invention. The present invention provides an impact load simulation system based on data restoration technology, comprising the following steps:

[0023] Step 1: Preprocess the collected load data;

[0024] Step 2: Process the preprocessed outliers;

[0025] Step 3: Establish an abnormal load data correction model;

[0026] Step 4: Analyze the data results after the model has been corrected.

[0027] Furthermore, in step one, the collected raw data is cleaned, missing values ​​are filled in, data with large missing values ​​are directly deleted, data with small missing values ​​are filled in using linear interpolation, and data with different dimensions are normalized.

[0028] If the variables of the load data are set as X and Y, then (x0, y0) and (x1, y1) are the corresponding data of two known X and Y values. The process of using linear interpolation to fill in the missing data is as follows:

[0029]

[0030] The variables have different dimensions, and the numerical differences between them within their respective dimensions are significant. Therefore, when processing data with different dimensions, we often need to perform data normalization, as shown in the following formula:

[0031]

[0032] Where P represents the original data in a certain dimension; P min P represents the minimum value of that dimension in the dataset. max This represents the maximum value of this dimension in the dataset. After normalizing the load data to the range [0,1], in order to effectively perform statistical analysis of the experimental results after the prediction work is completed, the results need to be denormalized to restore them to the original range of load data. The denormalization formula is as follows:

[0033] P=(P`)×(P max -P min )+P min (3)

[0034] Where P' is the normalized load data and P is the inverse normalized real data.

[0035] Furthermore, in step two, the preprocessed data is subjected to outlier detection using the 3σ principle, with the specific formula as follows:

[0036]

[0037]

[0038] Among them, P i,t This provides the raw load data for the user at time point i on a similar day, where n is the total number of similar days, and P... t and The average load and variance at time t on similar days are respectively;

[0039] Identifying abnormal numbers using the 3σ principle:

[0040]

[0041] Where ε is the threshold, which takes a value of 1 to 1.5;

[0042] After detecting outliers, outlier correction is performed using the following formula:

[0043]

[0044] Among them, P j,t Let t represent the load value at time t on day j, where j is the day number of the current time.

[0045] Furthermore, in step three, the preprocessed historical load data is divided into a training set, a test set, and a validation set for the anomaly data correction model, and the BP neural network prediction model is iteratively trained. The specific steps are as follows:

[0046] 1) Network initialization: Determine the number of input and output variables and provide the network connection structure; randomly initialize the weights of the connections between the input layer and the hidden layer, and between the hidden layer and the output layer.

[0047] 2) Provide training samples. If there are n input variables and m output variables, then each training sample is in the form of (x1, x2, ..., x...). n ;t1,t2,…,t m ), where t1, t2, ..., t m The input is x1, x2, ..., x n Expected output at that time;

[0048] 3) Calculate the actual output. Use a nonlinear function to calculate the output value of each node (excluding the input layer) level by level, and let the final output be (o1, o2, ..., o m The nonlinear function is as follows:

[0049]

[0050] 4) Weight adjustment: Use a recursive method to return from the output node to the hidden node. The formula for adjusting the weights is as follows:

[0051] ω ij (N+1)=ω ij (N)+ηo i δ j (9)

[0052] Among them, o i It is the output of the i-th node in the upper layer. If j is an output layer node, then

[0053] δj =o j (1-o j )(t j -o j (10)

[0054] If j is a hidden node, then

[0055]

[0056] Where k represents all nodes in the lower level of node j. Sometimes, to speed up convergence, an impulse term can be added to make the weights smoother, as shown below.

[0057] ω ij (N+1)=ω ij (N)+ηo i δ j +α[ω ij (N)-ω ij [(N-1)], 0<α<1 (12)

[0058] 5) Return to step 2 and repeat until the error is satisfactory.

[0059] See Figure 2 Furthermore, in step four, the mean absolute percentage error (MASE) is used to evaluate the prediction results. The closer the MSE is to 0, the better the prediction results.

[0060] The formula for the mean absolute percentage error is as follows:

[0061]

[0062] Where N represents the number of load samples, P ft P represents the predicted value at time t. rt The true value at time t is represented. The prediction error of the prediction model is evaluated through the test set. If the accuracy requirement is not met, the model parameters are adjusted through the validation set, and the process proceeds to step three. The relevant data before the occurrence of abnormal data is input into the trained BP neural network model for prediction. Finally, the predicted load value is output. In the abnormal data correction part, the predicted load value is used to replace the abnormal load value.

[0063] The above embodiments are implemented based on the technical solution of the present invention, providing detailed implementation methods and specific operation processes. However, the scope of protection of the present invention is not limited to the above embodiments. Unless otherwise specified, the methods used in the above embodiments are conventional methods.

Claims

1. An impact load simulation system based on data reconstruction technology, characterized in that, Includes the following steps: Step 1: Preprocess the collected load data; Step 2: Process outliers after preprocessing; use the 3σ principle to detect outliers in the preprocessed data, the specific formula is as follows: Among them, P i,t This provides the raw load data for the user at time point i on a similar day, where n is the total number of similar days, and P... t and The average load and variance at time t on similar days are respectively; Identifying abnormal numbers using the 3σ principle: Where ε is the threshold, which takes a value of 1 to 1.5; After detecting outliers, outlier correction is performed using the following formula: Among them, P j,t To represent the load value at time t on day j, where j is the day number of the current time; Step 3: Establish an abnormal load data correction model; divide the preprocessed historical load data into training, testing, and validation sets for the abnormal load data correction model, and iteratively train the BP neural network prediction model. The specific steps are as follows: 1) Network initialization: Determine the number of input and output variables and provide the network connection structure; randomly initialize the weights of the connections between the input layer and the hidden layer, and between the hidden layer and the output layer. 2) Provide training samples. If there are n input variables and m output variables, then each training sample is in the form of (x1, x2, ..., x...). n ;t1,t2,…,t m ), where t1, t2, ..., t m The input is x1, x2, ..., x n Expected output at that time; 3) Calculate the actual output. Use a nonlinear function to calculate the output value of each node at each level, excluding the input layer. Let the final output be (o1, o2, ..., o m The nonlinear function is as follows: 4) Weight adjustment: Use a recursive method to return from the output node to the hidden node. The formula for adjusting the weights is as follows: oh ij (N+1)=ω ij (N)+ηo i d j Among them, o i It is the output of the i-th node in the upper layer. If j is an output layer node, then δ j =o j (1-o j )(t j -o j ) If j is a hidden node, then Where k represents all nodes in the lower level of node j. To speed up convergence, an impulse term is added to make the weights smoother, as follows. oh ij (N+1)=ω ij (N)+ηo i d j +a[ω ij (N)-ω ij (N-1)],0<α<1 5) Return to step 2 and repeat until the error is satisfactory; Step 4: Analyze the data results after the model has been corrected.

2. The impact load simulation system based on data restoration technology according to claim 1, characterized in that, Step 1 cleans the collected raw data, fills in missing values, deletes data with large missing values, fills in data with small missing values ​​using linear interpolation, and normalizes data of different dimensions.

3. The impact load simulation system based on data restoration technology according to claim 1, characterized in that, Step four uses the mean absolute percentage error (MASE) evaluation index to measure the prediction results. The closer the MSE value is to 0, the better the prediction results.

Images